Use of Logic Relationships to Decipher Protein Network Organization Peter M. Bowers, Shawn J. Cokus, David Eisenberg, Todd O. Yeates

1
Use of Logic Relationshipsto Decipher
ProteinNetwork OrganizationPeter M. Bowers,
Shawn J. Cokus,David Eisenberg, Todd O. Yeates

• Presented by Krishna Balasubramanian

2
Contents

• Introduction
• Background
• Method Used - LAPP
• Results
• Observations
• Conclusion
• Future Work

3
Introduction

• Major focus of genome research
• Deciphering networks of molecular interactions
  underlying cellular function.
• Developed a Computational approach
• Identify detailed relationships btw proteins
  based on genomic data.
• The method reveals many previously unidentified
  higher order relationships

4
Background

• Patterns across multiple complete genomes have
  been used to infer biological interactions and
  functional linkages btw proteins
• 2 distinct proteins from one organism genetically
  fused into a single protein in another organism.
• Tendency of 2 proteins to occur in chromosomal
  proximity across multiple organisms.
• Phylogenetic profile approach
• Detects functional relationships btw proteins
  exhibiting statistically similar patterns of
  presence or absence.
• Determine pattern describing a proteins presence
  or absence by searching for its homologs across N
  organisms.

5
Background

• Original implementations sought to infer links
  btw pairs of proteins with similar profiles.
• A subsequent variation on that idea linked
  proteins if their profiles represented the
  negation of each other.
• Simple notions - with the presence of one protein
  implying the presence or absence of another.
• Such simple relationships cannot adequately
  describe the full complexity of cellular networks
  that involve branching, parallel, and alternate
  pathways.
• Higher order logic relationships involving a
  pattern of presence/absence of multiple proteins
  expected due to
• Observed complexity of cellular networks.
• Evolutionary divergence, convergence, and
  horizontal transfer events.

6
Method - LAPP

• Perform complete analysis of logic relations
  possible btw triplets of phylogenetic profiles.
• Demonstrate the power of the resulting logic
  analysis of phylogenetic profiles (LAPP) to
• Illuminate relationships among multiple proteins.
  
• Infer the coarse function of large numbers of
  uncharacterized protein families.

7

• Logical Relationships to determine
  presence/absence of Proteins
• Venn diagrams and logic statements show the 8
  distinct kinds of logic functions that describe
  the possible dependence of the presence of on the
  presence of A and B, jointly.
• Logic functions are grouped together if they are
  related by a simple exchange of proteins A and B.

8
Logical Relationships to determine
presence/absence of Proteins

• There are 8 possible logic relationships
  combining two phylogenetic profiles to match a
  third profile.
• E.g. 1 protein C might be present if and only if
  proteins A and B are both present.
• Function of protein C is necessary only when the
  functions of proteins A and B are both present.
• Gene C may be present if and only if either A or
  B is present.
• Different organisms use two different protein
  families in combination with a common third
  protein to accomplish some task.
• Several of the eight possible logic relationships
  intuitively understood to describe commonly
  observed biological scenarios.
• However, a few of the logic relationships are not
  easily related to real biological situations.

9
Examples of LAPP based on Phylogenetic
Profiles Phylogenetic Profiles Biological
examples of LAPP
10
Examples of LAPP .. Contd

• Hypothetical phylogenetic profiles are used to
  illustrate the eight possible logic functions.
• Real biological e.g. shown to illustrate the
  ternary relationships identified from actual
  phylogenetic profiles for the 4 most commonly
  observed logic types.

11
Identifying Protein Triplets

• Created a set of binary-valued vectors describing
  the presence or absence of each of the known
  protein families across 67 fully sequenced
  organisms.
• Categorized complete set of proteins into 4873
  distinct families called clusters of orthologous
  groups (COGs).
• Examined all triplet combinations of profiles and
  rank-ordered them according to how well the
  logical combination f (a,b) of two profiles
  predicted a third profile, c.
• Neither profile a nor b alone was predictive of c.

12
Identifying Protein Triplets

• Uncertainty Coefficients calculated for U(ca),
  U(cb), and the logically combined profile U(cf
  (a,b))
• U(xy) H(x) H(y) H(x, y)/H(x)
• H is the entropy of individual/joint
  distributions
• U can range between 1.0, where x is a
  deterministic function of y, and 0.0, where x is
  completely independent of y.
• Selected triplets whose individual pairwise
  uncertainty scores described protein profile c
  poorly U(ca) lt 0.3 and U(cb) lt 0.3 but whose
  logically combined profile U(cf (a,b)) gt 0.6
  described c well.

13
Example

• Synthesis of aromatic amino acids proceeds
  through the shikimate pathway.
• Logic analysis of 5 participating proteins show
• Shikimate can be converted to the end product
  prephenate by one of two possible routes, leading
  to a type 7 logic relationship.

Example showing triplet andpairwise
uncertaintycoefficients, U.
14
Results

• When either one shikimate kinase protein family
  (protein A, COG1685) or an alternate shikimate
  kinase protein family (protein B, COG0703) is
  present in an organism, then excitatory
  postsynaptic potential (EPSP) synthase must also
  be present (protein C, COG0128) (U 0 0.85) to
  carry out the subsequent enzymatic step.
• The same type 7 logic relationship is also
  observed between alternate shikimate kinase
  enzymes and the successive chorismate synthase
  (protein D, COG0082) and chorismate mutase
  (protein E, COG1605) enzymatic steps of the
  pathway.
• The ordering of the metabolic steps that follow
  shikimate kinase is predicted by the value of
  successive U coefficients, where EPSP synthase
  (second step, U 0 0.85) is most strongly linked
  to shikimate kinase, followed directly by the
  chorismate synthase (third step, U 0 0.66) and
  lastly by chorismate mutase (fourth step, U 0
  0.56).

15
Results Contd

• Organisms synthesize chorismate and prephenate
  from shikimate with the use of only one of two
  possible alternate routes pathways consisting of
  either ordered enzymes A-C-D-E or enzymes
  B-C-D-E.
• LAPP recovers 750,000 previously unknown
  relationships among protein families
  (U(c(f(a,b)) gt 0.60 U(cb) lt 0.30 U(ca) lt
  0.30).
• Validity assessed by comparing known annotations
  of the linked proteins.
• The ability to recover links between proteins
  annotated as belonging to a major functional
  category has been used widely to corroborate
  computational inferences of protein interactions.

16
Observations

• One of the most frequently observed triplet
  relationships relates three proteins belonging to
  the cell motility category, confirmation that the
  triplet associations link proteins closely
  related in function.
• Other triplets involve two proteins from the
  motility category and a third protein of another
  COG category, producing recognizable horizontal
  and vertical bands in the histogram.
• E.g. the category combinations NNU (COG category
  U, intracellular trafficking and secretion) and
  NNS (COG category S, unknown function) are also
  plentiful.
• Connections between these categories make
  intuitive sense and facilitate placement of
  unannotated proteins within the context of
  specific cellular networks of interacting
  proteins.

Section taken from a 3-D histogram that describes
the frequency of observed logic relationships in
which protein A of the triplet is annotated as
belonging to the COG functional category N, cell
motility.
17
Observations

• LAPP leads to a set of statistically significant
  ternary relationships that are distinct from and
  more numerous than the ones inferred using
  traditional pairwise analysis.
• Matrix of randomized phylogenetic profiles,
  containing the same individual and pairwise
  distributions as the native profiles used to
  assess the probability of observing a given
  uncertainty coefficient score by chance.
• Triplets with U gt 0.60 are observed from the
  unshuffled vectors 102 times more frequently
  than from shuffled profiles and 104 more
  frequently when U gt 0.80.

Plot of the cumulative number of protein triplets
recovered atan uncertainty coefficient score
greater than a given threshold.
18
Observations Contd

• P value calculated for each triplet relationship
  by enumerating all possible values of U that
  could be obtained from shuffled profiles while
  maintaining the individual and pairwise
  distributions.
• P number of trials that exceed the observed
  value of U divided by the total number of trials.
• More than 98 of the identified triplets (U gt
  0.6) have P lt 0.05, and more than 75 of the
  identified triplets have P lt 0.005.

19
Observations

• The 8 distinct logic types occur with widely
  varying frequencies within the set of significant
  ternary relationships.
• Consistent with our understanding of evolution
  biological relationships.
• Logic types 1, 3, 5, and 7 are observed
  frequently in the biological data.
• Logic types 2, 4, and 8 are more difficult to
  relate to simple cellular logic and are observed
  only rarely.

Number of identified triplets (U gt0.6) for each
of the eight logic function types for
randomized(black) and real (gray) phylogenetic
profiles.
20
Observations
50 highest scoring relationships (U gt 0.75)
involving proteins fromthe cell motility and
intracellular trafficking and secretion
functionalcategories.
21
Observations contd

• Cell motility proteins are colored light blue,
  intracellular trafficking and secretion are
  colored magenta, and proteins annotated as both
  are colored in orange.
• Edges are shown between proteins A-C and B-C of
  each logic triplet, with each edge labeled
  according to the logic function type used to
  associate the proteins families.

22
Observations contd

• The proteins linked include adhesin proteins
  necessary for bacterial pathogenesis, chemotaxis
  proteins, and translocase proteins.
• Network contains previously unknown interactions
  that suggest mechanisms connecting bacterial
  pathogenesis and chemotaxis.
• CheZ, a chemotaxis dephosphorylase that regulates
  cell motility, is linked to the surface receptor
  and virulence factors adhesin AidA and Flp
  pilus-associated FimT.

23
Conclusion

• New higher order protein associations detected by
  LAPP provides a framework to understand the
  complex logical dependencies that relate proteins
  to one another in the cell.
• Also useful in
• Modeling and engineering biological systems
• Generating biological hypotheses for
  experimentation
• Investigating additional protein properties

24
Future Work

• In all likelihood, logic relationships btw
  proteins in the cell extend beyond ternary
  relationships to include much larger sets of
  proteins.
• Ideas underlying the logical analysis of
  phylogenetic profiles can be extended to the
  investigation of other kinds of genomic data
• Gene expression,
• Nucleotide polymorphism
• Phenotype data

25
Questions??